惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

G
Google Developers Blog
有赞技术团队
有赞技术团队
WordPress大学
WordPress大学
博客园 - 司徒正美
D
Docker
B
Blog
V
Visual Studio Blog
Blog — PlanetScale
Blog — PlanetScale
U
Unit 42
S
SegmentFault 最新的问题
小众软件
小众软件
J
Java Code Geeks
美团技术团队
腾讯CDC
MyScale Blog
MyScale Blog
爱范儿
爱范儿
H
Help Net Security
宝玉的分享
宝玉的分享
Microsoft Azure Blog
Microsoft Azure Blog
A
About on SuperTechFans
Google DeepMind News
Google DeepMind News
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
博客园 - 三生石上(FineUI控件)
博客园 - 【当耐特】

Interesting Engineering

US firm to scale laser-based nuclear fusion ‘breakthrough’ with new partnership Military Archives - Interesting Engineering World’s first non-nuclear lead-cooled reactor to generate electricity begins installation US scientists devise new process to turn sewage sludge into 99% pure natural gas US firm unveils submarine-hunting drone with 9,200-mile-range, 35 mph top speed Military Archives - Interesting Engineering Supercomputer finds lithium-titanium tweak to boost sodium-ion batteries for grids Lockheed Martin demonstrates vertical launch missile system for mobile drone defense China’s 1116 MWe Taipingling Unit 1 reactor goes online, set to generate 9bn kWh yearly ChatGPT Images 2.0 update combines reasoning, research, and design with 2K output US Navy tests plug-and-play laser system on USS Bush carrier, downs drones at sea China’s CATL reveals 621-mile EV battery, under-7-minute charging to challenge BYD US uses world’s first exascale supercomputer to model supernovae, fusion reactors AI and Robotics Archives - Interesting Engineering First-in-human study confirms safety of graphene-based brain interface Tesla’s Optimus humanoid robot greets runners, poses for photos at Boston Marathon Interlocking materials offer high strength and flexibility for robotics, infrastructure US redeploys 100,000-ton nuclear-powered aircraft carrier in Red Sea after repairs US scientists unveil concept for ‘world’s first neutrino laser’ to unlock breakthroughs New military tech can maintain communication in contested electronic warfare environments Got a dark personality? Psychologists can help you choose your career wisely Humidity boosts performance of 3D-printed nanogenerator instead of degrading it China demonstrates microwave beam that recharges drones in flight, continues power delivery Scientists run compact free-electron laser for eight hours, cracks FEL stability problem China’s PLA considers to use minelaying underwater drones to enforce Taiwan blockade: Report 1-ton sharks may struggle for survival in waters exceeding 62.6°F, study suggests US firm’s thorium nuclear fuel bundles move to manufacturing for commercial reactors Tesla hits 0% charge in remote Chilean desert as YouTuber uses hood-mounted solar Humanoid robot surpasses human world record in Beijing half-marathon, clocking 50:26 mins New method extracts maximum work from unknown quantum states using symmetry tricks
SK hynix builds innovative cooling solution inside a 3D s...
Aditya Jadhav · 2026-05-28 · via Interesting Engineering

SK hynix has launched iHBM, an architecture featuring a Dynamic Random Access Memory (DRAM) stack with embedded Integrated Cooling Elements (ICEs). These cooling elements provide an additional heat dissipation pathway from the high-bandwidth memory (HBM) package. The ICEs are made of electrically non-conductive, thermally conductive silicon-based material.

Kangwook Lee, Senior Vice President and Head of Packaging Development at SK hynix said “iHBM is an optimal solution for thermal management, combining our memory design capabilities with advanced packaging technology”.

The thermal management problem

An HBM is a memory chip in which multiple DRAM dies are stacked vertically in a 3D layout. This shortens interconnect lengths, increase data speeds, reduce latency, and lower power consumption. Because of these benefits HBMs are heavily used in AI servers. The technology has evolved through successive generations from HBM, HBM2, HBM2E, HBM3, HBM3E, to HBM4.

For meeting the surging AI data processing demand, HBM technology is moving towards higher stacking and faster speeds. These two factors are also increasing the thermal load on the HBM package.

Moreover the key driver defining the competitiveness of next-generation HBM chips is the management of the thermal power density at an interface connecting HBM and the Graphic Processing Unit (GPU) also known as the Die-to-Die Physical Layer (D2D PHY). It is also a region with the highest heat concentration.

Architecture and integration

With the iHBM solution, SK hynix has taken a structural approach to addressing the thermal management problem. The existing HBM products rely on an indirect cooling method that draws heat away through the core die, whereas the iHBM solution places ICEs on top of the D2D PHY region. This creates an additional heat dissipation pathway to reduce the thermal load.

This heat management solution helps to reduce thermal resistance by 30 percent and enables chips to operate stably even in high-temperature and high-pressure conditions.

Industrial scale production

The company states that it has mass-production capabilities to enable stable high-volume production of iHBM-equipped chips (which are likely to be fabricated at its Cheongju M15X fab). These will utilize SK hynix`s Wafer Level Packaging (WLP) process based on Mass Reflow Molded Underfill (MR-MUF) technology. MR-MUF process enables stacking semiconductors by injecting liquid protective materials between chips.

The iHBM solution also offers high design compatibility with existing System-in-Package (SiP) architectures, allowing adoption to this new thermal management technology with minimal design adjustments.

The iHBM is, slated for deployment in next-generation HBM products, like the upcoming HBM5. With this product SK hynix aims to increase the stability and operational efficiency of AI data centers by meeting heat management standards required in high-density and high-bandwidth environments.

Future developments

SK hynix holds a dominant competitive position in HBM segment. It has also initiated proactive R&D in next-generation high-bandwidth storage technologies like High Bandwidth Flash (HBF) and Processing-In-Memory (PIM).

HBF is an emerging memory tier positioned between HBM and Solid-State Drive (SSD). It combines HBM class bandwidth characteristics with the high-density capacity of NAND flash-based storage, and is designed to improve scalability and power efficiency across AI infrastructure.

PIM is a next-generation memory architecture that integrates processing logic directly into the memory die, eliminating the traditional von Neumann bottleneck between memory and processor thereby dramatically improving bandwidth efficiency and energy consumption.

Recommended Articles